1 / 23

Artificial Intelligence Machine learning

Artificial Intelligence Machine learning . Fall 2008 professor: Luigi Ceccaroni. Knowledge acquisition. Reference frame: construction of a knowledge-based system: Knowledge acquisition Knowledge representation Definition of a resolution method Construction of an inference engine. 2.

efuru
Télécharger la présentation

Artificial Intelligence Machine learning

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Artificial Intelligence Machine learning Fall 2008 professor: Luigi Ceccaroni

  2. Knowledge acquisition • Reference frame: construction of a knowledge-based system: • Knowledge acquisition • Knowledge representation • Definition of a resolution method • Construction of an inference engine 2

  3. Base de coneixement Knowledge acquisition • Adquisició del coneixement ≡ Traspàs del coneixement d’un omés experts (ofonts de coneixement) en un dominideterminat, cap a un formalisme de representaciócomputable del coneixement FC1≡Expert1 . . . . FCn≡ Expertn Enginyer del coneixement Domini (fets, relacions, associacions) Procés de resolució (heurístiques, mètodes) Coneixement 3

  4. Methodologies to acquire knowledge • Interviews with experts • Techniques based on inductive machine learning • Construction of decision trees: • Nodes represent descriptors • Branches represent possible values of the descriptor 4

  5. Inductive machine learning • Techniques oriented to analysis problems (classification/interpretation) • The expert expresses his knowledge in a familiar form: observations/examples 5

  6. Inductive machine learning • Sample data: 6

  7. Inductive machine learning • First objective: Grouping similar objects via clustering methods 7

  8. Inductive machine learning • Second objective: Classifying new objects • The starting point is more informed, given that groups have been already defined. • Specific characteristics of each group/class are determined, to be able to assign a new object to a class. 8

  9. Inductive machine learning • Methods: • Decision trees: • CART, ID3, ASSISTANT, C4.5, C5.1 • Classification rules If Act1is steps Then Act2is ioga Rule's probability: 0.9 The rule exists in 52 records. 9

  10. Té febre Té tos Pols alterat Té mucositat Decision trees No A lot A little Yes No Yes No No Yes Decisión 2 Decisión 3 Decisión 4 Decisión 5 Decisión 6 Decisión 1

  11. ID3 • ID3 ≡ Induction of Decision Trees [Quinlan, 1979, 1986] • Top-down strategy • Starting from a set of examples/instances and their classes, it creates the tree which best explains them. 11

  12. ID3 It’s a greedy algorithm for the automatic construction of decision trees, which selects at each step the best descriptor. → The best one is the most discriminant (potentially more useful). It lets reduce the size of the decision tree. The selection is done maximizing a function G(X,A), which represents the information gain. 12

  13. ID3 • The construction process is iterative: • A subset of instances (window) is selected from the training set. • The decision tree is built, which allows to discriminate the instances of the window. • If the decision tree explains the rest of instances of the training set then it is the final tree, else the instances wrongly classified (exceptions) are added to the window and the process go back to (2) endIf

  14. Information gain • Information (X = instances, C = classification): • Entropy (A = attribute, [A(x) = vi ] represents instances with value vi ): • Information gain:

  15. ID3: example

  16. ID3: example I (X, C) = -1/2 log2 1/2 -1/2 log2 1/2 = 1 (1,2,5,8) (3,4,5,7) C+ C- E (X, Ulls) = 3/8 (-1 log2 1 - 0 log2 0) + 2/8 (-0 log2 0 - 1 log2 1) + 3/8 (-1/3 log2 1/3 - 2/3 log2 2/3) = 0.344 E (X, Cabell) = 2/8 (-1/2 log2 1/2 - 1/2 log2 1/2) + 6/8 (-3/6 log2 3/6 - 3/6 log2 3/6) = 1 E(X, Estatura) = 2/8 (-1 log2 1 - 0 log2 0) + 4/8 (-1/2 log2 1/2 - 1/2 log2 1/2) + 2/8 (-0 log2 0 - 1 log2 1) = 0.5 16

  17. Ulls 1,2,8 3,6 4,7, 5 + - - + ID3: example G (X, Ulls) = 1 - 0.344 = 0.656 G (X, Cabell) = 1 - 1 = 0 G (X, Estatura) = 1- 0.5 = 0.5 Verds Blaus Marrons Cabell Estatura Classe E4 Moreno Mitjà C- E5 Moreno Alt C+ E7 Ros Baix C-

  18. ID3:example I (X, C) = -1/3 log2 1/3 -2/3 log2 1/3 = 0,918 (5) (4,7) E (X, Cabell) = 1/3 (- 0 log2 0 - 1 log2 1) + 2/3 (-1/2 log2 1/2 - 1/2 log2 1/2) = 2/3 E (X, Estatura) = 1/3 (-0 log2 0 - 1 log2 1) + 1/3 (-1 log2 1 - 0 log2 0) + 1/3 (-0 log2 0 - 1 log2 1) = 0 G (X, Cabell) = 0,918 - 0,666 = 0,252 G (X, Estatura) = 0,918 - 0 = 0,918 Ros

  19. ID3:example Ulls Blaus Verds Marrons 1,2,8 3,6 4,7, 5 + - - + Estatura baix Alt Mitjà 5 4 7 + - -

  20. ID3:example Ulls = Blaus → C+ Ulls = Marrons → C- Ulls = Verds ∧ Estatura = Alt → C+ Ulls = Verds ∧ Estatura = Mitjà → C- Ulls = Verds ∧ Estatura = Baix → C-

  21. ID3: algorithm

  22. Tipos de aprendizaje • Aprendizaje inductivo: Creamos modelos de conceptos a partir de generalizar ejemplos simples. Buscamos patrones comunes que expliquen los ejemplos. • Aprendizaje analítico o deductivo: Aplicamos la deducción para obtener descripciones generales a partir de un ejemplo de concepto y su explicación.

  23. Tipos de aprendizaje • Aprendizaje genético: Aplica algoritmos inspirados en la teoría de la evolución para encontrar descripciones generales a conjuntos de ejemplos. • Aprendizaje conexionista: Busca descripciones generales mediante el uso de la capacidad de adaptación de redes de neuronas artificiales.

More Related